Vector2Variant: Discovery of Genetic Associations Without Phenotype Engineering (www.medrxiv.org)

A new framework called Vector2Variant (V2V) has been introduced to advance genome-wide association studies (GWAS) by eliminating the need for predefined phenotypes in genetic research. Unlike traditional approaches that link genetic variants to single traits, V2V utilizes high-dimensional measurements, such as machine learning embeddings, to identify the multidimensional axes where genetic carriers differ maximally from non-carriers. This innovation enables the generation of continuous "projection phenotypes" that correlate with various clinical biomarkers, leading to a more nuanced understanding of genetic associations tied to diseases. The significance of V2V in the AI and machine learning community lies in its ability to uncover genetic relationships without requiring targeted phenotype measurements. This method has been successfully applied to diverse data types, such as imaging, timeseries, and omics from the UK Biobank. Notably, V2V identified known genetic associations, like CASP9's role in renal failure, as well as novel links, such as a frameshift variant in LRRIQ1 that may provide cardiovascular protection. By efficiently processing data on a genome-wide scale, V2V not only streamlines genetic research but also enhances target prioritization in disease association studies.